- Researchers are developing machine learning techniques to enable uncrewed underwater vehicles (UUVs) to navigate autonomously without a Global Positioning System (GPS).
- These techniques involve deep reinforcement learning which involves random actions that are then observed and compared to a goal, with positive results reinforced and poor results avoided.
Underwater vehicles, or UUVs, have typically been used for tasks such as deep-sea exploration and disabling underwater mines. However, they often suffer from poor navigation control and communication due to water’s distorting effect. Traditional navigational techniques that use GPS or camera systems are ineffective underwater due to water’s obstruction of GPS signals and low visibility for cameras.
To address these challenges, a team of researchers from Australia and France are harnessing machine learning techniques to help UUVs navigate autonomously. The study, published in the IEEE Access journal, uses a type of machine learning called deep reinforcement learning.
This involves UUV models starting by making random actions, observing the results, then comparing them to the goal which, in this case, is to navigate as closely as possible to the target destination. Actions that lead to positive outcomes are reinforced, while less successful methods are discouraged.
Ocean currents provide another challenge for UUVs to overcome, adding unpredictability to their paths. The researchers have adjusted the conventional process of reinforcement learning to account for these conditions. They have altered the way the UUV samples from its memory buffer to mirror human learning methods more closely, focusing on recent experiences with large positive gains.
The researchers reported that UUV models trained via this adapted memory buffer technique improved more quickly and consumed less power, providing significant advantages when these vehicles are deployed. As such, they plan to test this new training algorithm on UUVs in the ocean soon.